Chris@10: /* Chris@10: * Copyright (c) 2003, 2007-11 Matteo Frigo Chris@10: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology Chris@10: * Chris@10: * This program is free software; you can redistribute it and/or modify Chris@10: * it under the terms of the GNU General Public License as published by Chris@10: * the Free Software Foundation; either version 2 of the License, or Chris@10: * (at your option) any later version. Chris@10: * Chris@10: * This program is distributed in the hope that it will be useful, Chris@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@10: * GNU General Public License for more details. Chris@10: * Chris@10: * You should have received a copy of the GNU General Public License Chris@10: * along with this program; if not, write to the Free Software Chris@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@10: * Chris@10: */ Chris@10: Chris@10: /* This file was automatically generated --- DO NOT EDIT */ Chris@10: /* Generated on Sun Nov 25 07:37:02 EST 2012 */ Chris@10: Chris@10: #include "codelet-dft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_notw_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -sign 1 -n 12 -name n1bv_12 -include n1b.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 48 FP additions, 20 FP multiplications, Chris@10: * (or, 30 additions, 2 multiplications, 18 fused multiply/add), Chris@10: * 49 stack variables, 2 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "n1b.h" Chris@10: Chris@10: static void n1bv_12(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT i; Chris@10: const R *xi; Chris@10: R *xo; Chris@10: xi = ii; Chris@10: xo = io; Chris@10: for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(24, is), MAKE_VOLATILE_STRIDE(24, os)) { Chris@10: V T1, T6, Tc, Th, Td, Te, Ti, Tz, T4, TA, T9, Tj, Tf, Tw; Chris@10: { Chris@10: V T2, T3, T7, T8; Chris@10: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: T6 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: T3 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@10: T7 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); Chris@10: T8 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: Tc = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: Th = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@10: Td = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: Te = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); Chris@10: Ti = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tz = VSUB(T2, T3); Chris@10: T4 = VADD(T2, T3); Chris@10: TA = VSUB(T7, T8); Chris@10: T9 = VADD(T7, T8); Chris@10: Tj = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: } Chris@10: Tf = VADD(Td, Te); Chris@10: Tw = VSUB(Td, Te); Chris@10: { Chris@10: V T5, Tp, TJ, TB, Ta, Tq, Tk, Tx, Tg, Ts; Chris@10: T5 = VADD(T1, T4); Chris@10: Tp = VFNMS(LDK(KP500000000), T4, T1); Chris@10: TJ = VSUB(Tz, TA); Chris@10: TB = VADD(Tz, TA); Chris@10: Ta = VADD(T6, T9); Chris@10: Tq = VFNMS(LDK(KP500000000), T9, T6); Chris@10: Tk = VADD(Ti, Tj); Chris@10: Tx = VSUB(Tj, Ti); Chris@10: Tg = VADD(Tc, Tf); Chris@10: Ts = VFNMS(LDK(KP500000000), Tf, Tc); Chris@10: { Chris@10: V Tr, TF, Tb, Tn, TG, Ty, Tl, Tt; Chris@10: Tr = VADD(Tp, Tq); Chris@10: TF = VSUB(Tp, Tq); Chris@10: Tb = VSUB(T5, Ta); Chris@10: Tn = VADD(T5, Ta); Chris@10: TG = VADD(Tw, Tx); Chris@10: Ty = VSUB(Tw, Tx); Chris@10: Tl = VADD(Th, Tk); Chris@10: Tt = VFNMS(LDK(KP500000000), Tk, Th); Chris@10: { Chris@10: V TC, TE, TH, TL, Tu, TI, Tm, To; Chris@10: TC = VMUL(LDK(KP866025403), VSUB(Ty, TB)); Chris@10: TE = VMUL(LDK(KP866025403), VADD(TB, Ty)); Chris@10: TH = VFNMS(LDK(KP866025403), TG, TF); Chris@10: TL = VFMA(LDK(KP866025403), TG, TF); Chris@10: Tu = VADD(Ts, Tt); Chris@10: TI = VSUB(Ts, Tt); Chris@10: Tm = VSUB(Tg, Tl); Chris@10: To = VADD(Tg, Tl); Chris@10: { Chris@10: V TK, TM, Tv, TD; Chris@10: TK = VFMA(LDK(KP866025403), TJ, TI); Chris@10: TM = VFNMS(LDK(KP866025403), TJ, TI); Chris@10: Tv = VSUB(Tr, Tu); Chris@10: TD = VADD(Tr, Tu); Chris@10: ST(&(xo[0]), VADD(Tn, To), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 6)]), VSUB(Tn, To), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 9)]), VFMAI(Tm, Tb), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 3)]), VFNMSI(Tm, Tb), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 5)]), VFMAI(TM, TL), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 7)]), VFNMSI(TM, TL), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 11)]), VFNMSI(TK, TH), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 1)]), VFMAI(TK, TH), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 8)]), VFNMSI(TE, TD), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 4)]), VFMAI(TE, TD), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 2)]), VFMAI(TC, Tv), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 10)]), VFNMSI(TC, Tv), ovs, &(xo[0])); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 12, XSIMD_STRING("n1bv_12"), {30, 2, 18, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n1bv_12) (planner *p) { Chris@10: X(kdft_register) (p, n1bv_12, &desc); Chris@10: } Chris@10: Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_notw_c.native -simd -compact -variables 4 -pipeline-latency 8 -sign 1 -n 12 -name n1bv_12 -include n1b.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 48 FP additions, 8 FP multiplications, Chris@10: * (or, 44 additions, 4 multiplications, 4 fused multiply/add), Chris@10: * 27 stack variables, 2 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "n1b.h" Chris@10: Chris@10: static void n1bv_12(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT i; Chris@10: const R *xi; Chris@10: R *xo; Chris@10: xi = ii; Chris@10: xo = io; Chris@10: for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(24, is), MAKE_VOLATILE_STRIDE(24, os)) { Chris@10: V T5, Ta, TG, TF, Ty, Tm, Ti, Tp, TJ, TI, Tx, Ts; Chris@10: { Chris@10: V T1, T6, T4, Tk, T9, Tl; Chris@10: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: T6 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: { Chris@10: V T2, T3, T7, T8; Chris@10: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: T3 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@10: T4 = VADD(T2, T3); Chris@10: Tk = VSUB(T2, T3); Chris@10: T7 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); Chris@10: T8 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: T9 = VADD(T7, T8); Chris@10: Tl = VSUB(T7, T8); Chris@10: } Chris@10: T5 = VFNMS(LDK(KP500000000), T4, T1); Chris@10: Ta = VFNMS(LDK(KP500000000), T9, T6); Chris@10: TG = VADD(T6, T9); Chris@10: TF = VADD(T1, T4); Chris@10: Ty = VADD(Tk, Tl); Chris@10: Tm = VMUL(LDK(KP866025403), VSUB(Tk, Tl)); Chris@10: } Chris@10: { Chris@10: V Tn, Tq, Te, To, Th, Tr; Chris@10: Tn = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tq = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@10: { Chris@10: V Tc, Td, Tf, Tg; Chris@10: Tc = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: Td = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); Chris@10: Te = VSUB(Tc, Td); Chris@10: To = VADD(Tc, Td); Chris@10: Tf = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tg = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: Th = VSUB(Tf, Tg); Chris@10: Tr = VADD(Tf, Tg); Chris@10: } Chris@10: Ti = VMUL(LDK(KP866025403), VSUB(Te, Th)); Chris@10: Tp = VFNMS(LDK(KP500000000), To, Tn); Chris@10: TJ = VADD(Tq, Tr); Chris@10: TI = VADD(Tn, To); Chris@10: Tx = VADD(Te, Th); Chris@10: Ts = VFNMS(LDK(KP500000000), Tr, Tq); Chris@10: } Chris@10: { Chris@10: V TH, TK, TL, TM; Chris@10: TH = VSUB(TF, TG); Chris@10: TK = VBYI(VSUB(TI, TJ)); Chris@10: ST(&(xo[WS(os, 3)]), VSUB(TH, TK), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 9)]), VADD(TH, TK), ovs, &(xo[WS(os, 1)])); Chris@10: TL = VADD(TF, TG); Chris@10: TM = VADD(TI, TJ); Chris@10: ST(&(xo[WS(os, 6)]), VSUB(TL, TM), ovs, &(xo[0])); Chris@10: ST(&(xo[0]), VADD(TL, TM), ovs, &(xo[0])); Chris@10: } Chris@10: { Chris@10: V Tj, Tv, Tu, Tw, Tb, Tt; Chris@10: Tb = VSUB(T5, Ta); Chris@10: Tj = VSUB(Tb, Ti); Chris@10: Tv = VADD(Tb, Ti); Chris@10: Tt = VSUB(Tp, Ts); Chris@10: Tu = VBYI(VADD(Tm, Tt)); Chris@10: Tw = VBYI(VSUB(Tt, Tm)); Chris@10: ST(&(xo[WS(os, 11)]), VSUB(Tj, Tu), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 5)]), VADD(Tv, Tw), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 1)]), VADD(Tj, Tu), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 7)]), VSUB(Tv, Tw), ovs, &(xo[WS(os, 1)])); Chris@10: } Chris@10: { Chris@10: V Tz, TD, TC, TE, TA, TB; Chris@10: Tz = VBYI(VMUL(LDK(KP866025403), VSUB(Tx, Ty))); Chris@10: TD = VBYI(VMUL(LDK(KP866025403), VADD(Ty, Tx))); Chris@10: TA = VADD(T5, Ta); Chris@10: TB = VADD(Tp, Ts); Chris@10: TC = VSUB(TA, TB); Chris@10: TE = VADD(TA, TB); Chris@10: ST(&(xo[WS(os, 2)]), VADD(Tz, TC), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 8)]), VSUB(TE, TD), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 10)]), VSUB(TC, Tz), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 4)]), VADD(TD, TE), ovs, &(xo[0])); Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 12, XSIMD_STRING("n1bv_12"), {44, 4, 4, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n1bv_12) (planner *p) { Chris@10: X(kdft_register) (p, n1bv_12, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */